PS2 question

[Fafalada]

Actually, I found your last statement a bit ironic given someone else wrote "And don't dig up those old 4- and 8-bit pieces of art - they had VERY limited colour ranges". It just shows that some people clearly see such 'near monochromatic' textures as somewhat bland. <shrug>

Oh come now. 2x2VQ comes with a "built in" dither due to nature of compression itself, as well as looking worse then 8bit at same dimensions - either way it's anything but rich in color. (Especially when using 16bit entries in your codebook) Yet I'd bet some of the same people crying about limited 8bit palettes would argue the exact opposite for 2x2VQ. :?

Not that it's too surprising - more then once I've seen examples of people bitching about low quality of 8bit textures, using actual 4bit maps as ahem... "proof"...

 

[Simon F]

Actually, I found your last statement a bit ironic given someone else wrote "And don't dig up those old 4- and 8-bit pieces of art - they had VERY limited colour ranges". It just shows that some people clearly see such 'near monochromatic' textures as somewhat bland. <shrug>

Oh come now. 2x2VQ comes with a "built in" dither due to nature of compression itself, as well as looking worse then 8bit at same dimensions - either way it's anything but rich in color. (Especially when using 16bit entries in your codebook) Yet I'd bet some of the same people crying about limited 8bit palettes would argue the exact opposite for 2x2VQ. :?

Not that it's too surprising - more then once I've seen examples of people bitching about low quality of 8bit textures, using actual 4bit maps as ahem... "proof"...

Don't get the "wrong end of the stick". I didn't say 2bpp VQ was better than an 8bpp CLUT!

I said that 4bpp CLUT was worse than 4bpp S3TC, and that quality-wise, it was about the same as 2bpp VQ.

 

[Fafalada]

Sorry I guess I kind of jumped on the comment you only quoted from someone else (doesn't help that I'm a bit tired too).

Anyway yeah, 4bit is quite weak outside hand created textures, I'm still looking for a decent alternative. (luminance compression could be it if it weren't for mipmapping problems with it...)

 

[marconelly!]

I would think that the RMS errors and RMS error/bit factors that I gave in comparison are quite significant.

True, but I was just going by visual comparision.

Besides, I think it's safe to say pictures like these for example:

Have colour ranges bigger than 99% of textures that are normally used.

 

[Thowllly]

Now I just have to post (one of) my favourite 8bit picture(s):

mmmm...

 

[chaphack]

I dont see the point of posting 8bit pics. I mean, theoretically PS2 can do nice things, but realworld....unlesss we start to see something nice....

Come nextgen, 8bit textures will be null and void.

 

[andypski]

Now I just have to post (one of) my favourite 8bit picture(s):
mmmm...

Mmmm.... dithering.

[Sarcasm on]
I always think it's amazing how good dithering looks on textures when they're blown up and bilinearly filtered.
[Sarcasm off]



Some methods of making 8 bit images look better on monitors at 1:1 just don't work very well with 3D images.

It is a nice colourful picture though.

 

[marconelly!]

I dont see the point of posting 8bit pics.

The point is to show people that 256 colours selected from a 16M palette is quite enough to make a vibrantly coloured picture, as many seem to think that is the reason for 'washed out' colors (which are more often than not artist's choice, in todays market where games with colors that poke your eyes out are simply not as appreciated anymore)

I mean, theoretically PS2 can do nice things, but realworld....unlesss we start to see something nice....

Are you implying there aren't nicely textured games on PS2? That's just ridiculous...

 

[Simon F]

The point is to show people that 256 colours selected from a 16M palette is quite enough to make a vibrantly coloured picture, as many seem to think that is the reason for 'washed out' colors (which are more often than not artist's choice, in todays market where games with colors that poke your eyes out are simply not as appreciated anymore)

But it's a different issue with 4bpp palettised textures. 16 colours are very limiting and hence you can get either 'washed out' or rather dithered results.

I thought we were discussing the relative merits of texture compression schemes - clearly if you throw more bits at a texture you can achieve better quality but you have to pay for it.

 

[andypski]

The point is to show people that 256 colours selected from a 16M palette is quite enough to make a vibrantly coloured picture, as many seem to think that is the reason for 'washed out' colors (which are more often than not artist's choice, in todays market where games with colors that poke your eyes out are simply not as appreciated anymore)

I thought we were discussing the relative merits of texture compression schemes - clearly if you throw more bits at a texture you can achieve better quality but you have to pay for it.

And in the case of 8-bpp palettised vs. 4-bpp DXTC you throw more bits at it and still almost always end up with a worse result. Just about the only time you can do better with palettes is with a monochromatic image - if it's vibrantly coloured DXTC will almost always do a better job.

Even if the texture only has fairly mild variations in base chroma DXTC will still do better because the palette rapidly has to start trading colour variation for its resolution in luminance variation.

 

[Simon F]

Apologies to the artist for re-using the image but...

Now I just have to post (one of) my favourite 8bit picture(s):

mmmm...

Just to see how that copes with the various alternatives, I've compressed that image in various ways:

<Arghh changed these to indirect links for those with slow connections!>

A 4bpp CLUT version (using gimps quantiser):
4BPP CLUT

4bpp S3TC:

The 2bpp VQ (which is really struggling with this image):

Interestingly, the VQ compressor has detected the hidden watermark I suspect this means that my compressor code could have been better. <shrug>

 

[chaphack]

I dont know but....the CLUT jpeg looks really depressing.

 

[Simon F]

I dont know but....the CLUT jpeg looks really depressing.

Then it's a good thing you didn't see the palette that another application selected 'as optimal'!

 

[Phil]

Upon examening the 4 pictures size, I found something interesting:

- 8 BPP CLUT: 265264 bytes
- 4 BPP CLUT: 363968 bytes (WTF?)
- 4 BPP S3TC: 240203 bytes
- 2 BPP VQ: 255013 bytes

Besides, when using paletised colours, I guess you can always try to slice them in a way to get the maximum out of the given part and the available colours...

 

[Thowllly]

Upon examening the 4 pictures size, I found something interesting:

- 8 BPP CLUT: 265264 bytes
- 4 BPP CLUT: 363968 bytes (WTF?)
- 4 BPP S3TC: 240203 bytes
- 2 BPP VQ: 255013 bytes

Well, what is interesting about it? They are jpg compressed. If they hadn't been converted back to 24bpp and saved as a jpg they would have had the following sizes, without additional compression:
- 8 BPP CLUT: 321kB
- 4 BPP CLUT: 160kB
- 4 BPP S3TC: 160kB
- 2 BPP VQ: 82kB (assuming 16bit color)

Anyway, Simon, I think you should have used PNG, I used Gimp myself to reduce the image to 16 colors, and when saved as a PNG the image only took 94KB.

 

[Phil]

oops, you're right. I probably just looked at the 8bit CLUT filename (GIF), leaving out the other ones which were obviously changed back to 24 bit. Damn, just too busy at work... :?

 

[andypski]

Simon - did you use your DXTC compressor for this test? It guess it did pretty well, although it's difficult to tell - the JPG artifacts are rather more visible than the ones caused by the DXTC conversion - 'free' dithering

If only we had some original artwork for this in 24bpp...

 

[Simon F]

Anyway, Simon, I think you should have used PNG, I used Gimp myself to reduce the image to 16 colors, and when saved as a PNG the image only took 94KB.

DOH! I tried PNG (24bit) for the other 2 formats and found it was a bit larger than "Max quality" JPG and so just stuck with the latter. I'd forgotten that PNG has a palette mode as well.

Simon - did you use your DXTC compressor for this test? It guess it did pretty well, although it's difficult to tell

No that's the S3 one. I don't know what I've done with my old code. :?

the JPG artifacts are rather more visible than the ones caused by the DXTC conversion - 'free' dithering

Well, I was just trying to show the differences between 4bit CLUT and the alternatives, and the error due to JPG seemed relatively 'insignificant'. Also I'd originally intended to display them directly, but they ended up being a bit lardy.
In retrospect perhaps I should change them to PNG.

If only we had some original artwork for this in 24bpp...

Perhaps if we put them through a good filter we'd average out the dithering and get an approximation of a 24bpp image?

 

[Thowllly]

I converted the original gif to a s3tc, converted it to 24bit, saved it as a PNG, and reduced the size of it with Pngcrush:
LAZUR-Fairylands3tc.png 318kB.
Looks good.

edit: oh, never mind, Simon has allready done it himself...

But my file is smaller

 

[Simon F]

Ok they are now all PNGs. I suspect this will blow the monthly bandwidth budget on that site